An AI-based experimental equipment data processing method

Abstract

The present invention relates to the field of data processing, and more specifically, to an experimental equipment data processing method based on AI, the method comprising: obtaining a time series sequence of experimental equipment data in each dimension; obtaining the spectrum of the experimental equipment data time series sequence, arranging all frequency components in the spectrum in ascending order according to the frequency size, superimposing the frequency components in sequence, calculating the regular change amount after each superposition of a new frequency component, and screening out a number of suspicious components by comparing the regular change amount with a preset change amount threshold; calculating the noise degree of the suspicious components, clustering the suspicious components into two categories according to the noise degree, removing the suspicious components in the category with a large noise degree mean, and realizing denoising. The accuracy of denoising is improved by accurately extracting the frequency components corresponding to the noise.

Description

Technical Field

[0001] The present invention relates to the field of data processing, and more specifically, to an AI-based experimental equipment data processing method. Background Art

[0002] The performance and operation status of the experimental equipment are the basis for ensuring the effective operation of the experiment, so the performance and operation status of the experimental equipment need to be monitored during the experiment. In the process of monitoring the performance and operation status of the experimental equipment, the experimental equipment data will be collected. Due to electromagnetic or other factors, there is noise in the collected experimental equipment data, which will affect the analysis results. Therefore, before using the experimental equipment data for relevant analysis, denoising processing is required.

[0003] Fourier transform algorithm is often used for denoising. In the process of denoising using Fourier transform algorithm, noise is often considered as high-frequency component, so denoising is achieved by removing high-frequency components. However, due to the large frequency of some experimental equipment data, it is easy to remove useful information in the experimental equipment data when removing high-frequency components. Therefore, the key to denoising using Fourier transform algorithm is how to accurately filter out the components corresponding to the noise.

[0004] The patent application document with publication number CN109614681A discloses a method for removing trend from spectral data based on Fourier transform and wavelet analysis. The method in the patent application document only obtains the frequency range through the Fourier transform algorithm, and then sets the number of wavelet decomposition layers according to the frequency range. Therefore, the method in the patent application document does not involve relevant content on how to accurately screen out the corresponding components of the noise. Therefore, the method in the patent application document cannot solve the technical problem of the present invention. Summary of the invention

[0005] In order to solve the problem of how to accurately filter out the components corresponding to the noise, the present invention proposes an experimental equipment data processing method based on AI, which includes the following steps:

[0006] Obtain the experimental equipment data time series for each dimension;

[0007] Obtain the spectrum of the experimental equipment data time series, arrange all frequency components in the spectrum in ascending order according to the frequency size, superimpose the frequency components in sequence, and calculate the regularity change amount after each superposition of a new frequency component , obtain the newly added peaks / valleys after superimposing the new frequency components and the monotonic segment where each newly added peak / valley is located, Indicates the first The slope at the data point, Indicates the number of data in the experimental equipment data time series sequence, , Respectively represent the maximum fluctuation and span of the jth newly added peak / valley, represents the length of the monotonic segment where the jth newly added peak / valley is located, M represents the number of newly added peaks / valleys, G represents the regular change amount, norm() represents linear normalization processing, and a number of suspicious components are screened out by comparing the regular change amount with the preset change amount threshold;

[0008] Calculate the noise level of the suspected component , X represents the mean correlation between the suspicious component and the experimental equipment time series of all other dimensions, They respectively represent the distribution interval difference and morphological difference between the newly added peaks obtained after superimposing new suspicious components. The suspicious components are clustered into two categories according to the noise level, and the suspicious components in the category with a large mean noise level are removed to achieve denoising.

[0009] The present invention selects the frequency components corresponding to the noise signal according to the characteristics of the noise to achieve more accurate denoising processing; further, considering the characteristic that the change amount of the noise signal to the signal regularity is small, the frequency components corresponding to the noise are preliminarily selected by analyzing the regular change amount caused by each frequency component; further, considering that the change amount of the abnormal signal to the signal regularity is also small, the abnormal signal can also be extracted through the regular change amount, and thus the interference of the abnormal signal is eliminated by introducing the noise degree analysis, so that the frequency components corresponding to the noise are more accurately extracted; further, when analyzing the regular change amount, the slope, the maximum fluctuation amount and the span of the newly added peak are introduced to more comprehensively and accurately reflect the change of the regularity caused by the addition of the frequency component; further, when analyzing the noise degree, the noise and abnormal signal are more accurately distinguished by introducing the correlation, the distribution interval difference and the morphology difference.

[0010] Preferably, the processing of sequentially superimposing the frequency components includes:

[0011] First, the first frequency component is taken as the reference sequence, and the sequence obtained by superimposing the second frequency component and the reference sequence is recorded as a new reference sequence; then the sequence obtained by superimposing the third frequency component and the new reference sequence is recorded as a new reference sequence, and so on, until all frequency components are superimposed.

[0012] Preferably, the step of obtaining the newly added peaks / valleys after the new frequency components are superimposed and the monotonic segment where each newly added peak / valley is located includes:

[0013] The sequence after the new frequency component is superimposed is recorded as the post-superimposition sequence, and the sequence before the new frequency component is superimposed is recorded as the pre-superimposition sequence;

[0014] Get all extreme values ​​in the sequence before superposition, and get all extreme values ​​in the sequence after superposition;

[0015] Compared with the sequence before superposition, the extreme value that only exists in the sequence after superposition is taken as the target extreme value;

[0016] Get the extreme values ​​on both sides of the target extreme value, and use the area between the extreme values ​​on both sides of the superimposed sequence as the newly added peak / valley;

[0017] The sequence before superposition is divided into several monotonic segments using extreme points; the time period corresponding to the monotonic segment is obtained, the time period corresponding to the newly added peak / valley is obtained, and the monotonic segment that intersects with the time period of the newly added peak / valley is taken as the monotonic segment where the newly added peak / valley is located.

[0018] The present invention extracts the newly added peaks / valleys by comparing and analyzing the signals before and after the frequency components are superimposed. This type of processing is relatively simple and has a higher implementation efficiency.

[0019] Preferably, the method for obtaining the maximum fluctuation amount includes:

[0020] The two boundary points of the newly added peak / valley are obtained, and the average of the extreme value in the newly added peak / valley and the longitudinal distance between the two boundary points is taken as the maximum fluctuation amount of the newly added peak / valley.

[0021] The present invention measures the change of the signal trend before superposition caused by the newly added peaks / valleys by analyzing the relative height difference between the extreme point in the newly added peaks / valleys and the boundary points on both sides. This measurement method is more accurate and has higher implementation efficiency.

[0022] Preferably, the method for obtaining the span includes:

[0023] The two boundary points of the newly added peak / valley are obtained, and the horizontal distance between the two boundary points of the newly added peak / valley is used as the span of the newly added peak / valley.

[0024] Preferably, the method for obtaining the distribution interval difference includes:

[0025] The horizontal distance between the extreme point of the newly added peak / valley and the extreme point of the previous newly added peak / valley is obtained as the distribution interval of the newly added peak / valley, and the variance of the distribution intervals of all the newly added peaks / valleys is taken as the distribution interval difference.

[0026] The present invention measures the distribution regularity of newly added peaks / valleys more accurately by analyzing the time interval differences between adjacent newly added peaks / valleys, thereby providing a data basis for accurately eliminating abnormal signals.

[0027] Preferably, the method for obtaining the morphological difference comprises:

[0028] The cosine similarity between the newly added peak / valley and the previous newly added peak / valley is obtained as the adjacent similarity of the newly added peak / valley, and the average of the adjacent similarities of all the newly added peaks / valleys is taken as the morphological difference.

[0029] The present invention measures the changing regularity of the newly added peaks / valleys more accurately by analyzing the morphological differences between adjacent newly added peaks / valleys, thereby providing a data basis for accurately eliminating abnormal signals.

[0030] Preferably, removing suspicious components in the category with a large noise level mean value to achieve denoising includes:

[0031] The mean of the noise levels of all suspicious components in each category is calculated, the suspicious components in the category with the mean noise level are removed, and all the remaining frequency components are superimposed together to obtain the denoised experimental equipment data time series.

[0032] The present invention has the following beneficial effects:

[0033] The present invention selects the frequency components corresponding to the noise signal according to the characteristics of the noise to achieve more accurate denoising processing;

[0034] Furthermore, considering the fact that the noise signal has a small change in the signal pattern, the frequency components corresponding to the noise are preliminarily screened out by analyzing the change in the pattern caused by each frequency component;

[0035] Furthermore, considering that the change of the signal regularity by the abnormal signal is also small, the abnormal signal can also be extracted through the change of the regularity. Therefore, by introducing the noise degree analysis to eliminate the interference of the abnormal signal, the frequency component corresponding to the noise can be extracted more accurately.

[0036] Furthermore, when analyzing the change in the law, the slope, the maximum fluctuation and span of the newly added peak are introduced to more comprehensively and accurately reflect the change in the law caused by the addition of the frequency component.

[0037] Furthermore, when analyzing the noise level, the noise and abnormal signals can be distinguished more accurately by introducing correlation, distribution interval difference and morphology difference. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] By reading the following detailed description with reference to the accompanying drawings, the above and other objects, features and advantages of the exemplary embodiments of the present invention will become readily understood. In the accompanying drawings, several embodiments of the present invention are shown in an exemplary and non-limiting manner, and the same or corresponding reference numerals represent the same or corresponding parts, wherein:

[0039] Figure 1 It is a step flow chart of an AI-based experimental equipment data processing method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0040] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0041] The specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0042] See also Figure 1 , which shows a flowchart of a method for processing experimental equipment data based on AI according to an embodiment of the present invention, the method comprising the following steps:

[0043] S1: Obtain the time series of experimental equipment data in each dimension.

[0044] Specifically, sensors are used to collect data of each dimension of the experimental equipment at each moment, and the data of all moments in each dimension are arranged in time series to obtain a time series sequence of experimental equipment data in each dimension.

[0045] Among them, the experimental equipment can be a temperature controller, and the dimension type can be current, voltage, temperature, etc.

[0046] S2: Obtain the spectrum of the experimental equipment data time series sequence; arrange all frequency components in the spectrum in ascending order according to the frequency size, superimpose the frequency components in sequence, and calculate the regular change after each superposition of new frequency components; screen out several suspicious components by comparing the regular change amount with a preset change amount threshold.

[0047] S20: Acquire the frequency spectrum of the experimental equipment data time series sequence.

[0048] Preferably, as an example, obtaining the frequency spectrum of the experimental equipment data time series sequence includes:

[0049] The experimental equipment data time series sequence is processed by using a Fourier transform algorithm to obtain a frequency spectrum of the experimental equipment data time series sequence.

[0050] S21: Arrange all frequency components in the spectrum in ascending order according to the frequency magnitude, superimpose the frequency components in sequence, and calculate the regularity change amount after superimposing a new frequency component each time.

[0051] It should be noted that the noise signal has a small content, low intensity and no regularity. The addition of the noise signal will not cause much change in the signal regularity. Therefore, the situation of each frequency component being noise can be judged by analyzing the impact of the addition of each frequency component on the regularity.

[0052] Preferably, as an example, all frequency components in the spectrum are arranged in ascending order according to the frequency magnitude, the frequency components are sequentially superimposed, and the regularity change amount is calculated after each superposition of a new frequency component, including:

[0053] Arrange all frequency components in the spectrum in ascending order according to frequency to obtain a frequency component sequence;

[0054] First, the first frequency component is taken as the reference sequence, and the sequence obtained by superimposing the second frequency component and the reference sequence is recorded as a new reference sequence; then the sequence obtained by superimposing the third frequency component and the new reference sequence is recorded as a new reference sequence, and so on, until all frequency components are superimposed.

[0055] Calculate the change in the regularity after each superposition of a new frequency component:

[0056]

[0057] Among them, the newly added peaks / valleys after superimposing the new frequency components and the monotonic segment where each newly added peak / valley is located are obtained. Indicates the first The slope at the data point, Indicates the first The slope at the data point, It represents the number of data in the experimental equipment data time series, which is the same as the number of data in the corresponding sequence before and after the superposition of the new frequency component. represents the maximum fluctuation of the jth newly added peak / valley, represents the span of the jth newly added peak / valley, It represents the length of the monotonic segment where the jth newly added peak / valley is located, M represents the number of newly added peaks / valleys, G represents the amount of change in the regularity, and norm() represents linear normalization processing.

[0058] Understandably, It reflects the trend change. The larger the value, the greater the trend change caused by the addition of this frequency component. Since the noise content is small, the intensity is weak and the regularity is weak, the addition of noise will not change the trend law of the signal to a large extent. Therefore, if the addition of this frequency component causes a large change in the trend law of the signal, the possibility that this frequency component is noise is small. It reflects the blocking of the original monotonic law by the newly added peaks / valleys. The larger the value, the greater the change of the original monotonic law by the newly added peaks / valleys. Therefore, the possibility that the frequency component is noise is smaller. It reflects the change of the overall trend law caused by the addition of this frequency component. It reflects the change of local trend law caused by the addition of this frequency component. It can more comprehensively reflect the changes in trend patterns caused by the addition of this frequency component.

[0059] The above embodiments involve newly added peaks / valleys and the monotonic segment, maximum fluctuation and span of each newly added peak / valley. The following describes a method for obtaining the newly added peaks / valleys and the monotonic segment, maximum fluctuation and span of each newly added peak / valley.

[0060] First, a method for obtaining the newly added peaks / valleys and the monotonic segment where each newly added peak / valley is located is introduced.

[0061] Preferably, as an example, a method for acquiring the newly added peaks / valleys and the monotonic segment where each newly added peak / valley is located includes:

[0062] The sequence after the new frequency component is superimposed is recorded as the post-superimposition sequence, and the sequence before the new frequency component is superimposed is recorded as the pre-superimposition sequence;

[0063] Get all extreme values ​​in the sequence before superposition, and get all extreme values ​​in the sequence after superposition;

[0064] Compared with the sequence before superposition, the extreme value that only exists in the sequence after superposition is taken as the target extreme value;

[0065] Get the extreme values ​​on both sides of the target extreme value, and use the area between the extreme values ​​on both sides of the superimposed sequence as the newly added peak / valley;

[0066] The sequence before superposition is divided into several monotonic segments using extreme points; the time period corresponding to the monotonic segment is obtained, the time period corresponding to the newly added peak / valley is obtained, and the monotonic segment that intersects with the time period of the newly added peak / valley is taken as the monotonic segment where the newly added peak / valley is located.

[0067] Then the method of obtaining the maximum fluctuation and span is introduced.

[0068] Preferably, as an example, the method for obtaining the maximum fluctuation amount and span includes:

[0069] The two boundary points of the newly added peak / valley are obtained, and the average of the extreme value in the newly added peak / valley and the longitudinal distance between the two boundary points is taken as the maximum fluctuation amount of the newly added peak / valley.

[0070] The two boundary points of the newly added peak / valley are obtained, and the horizontal distance between the two boundary points of the newly added peak / valley is used as the span of the newly added peak / valley.

[0071] It can be understood that the maximum fluctuation amount and span can reflect the intensity of the regular changes caused by the newly added peaks / valleys.

[0072] S22: Screen out a number of suspicious components by comparing the regular change amount with a preset change amount threshold.

[0073] Preferably, as an example, a number of suspicious components are screened out by comparing the regular change amount with a preset change amount threshold, including:

[0074] The frequency component whose regular change amount is less than the preset change amount threshold is regarded as a suspicious component. This embodiment is described by taking the preset change amount threshold of 0.2 as an example. Other embodiments may take other values, and this embodiment does not make specific limitations.

[0075] S3: Calculate the noise level of the suspicious components; cluster the suspicious components into two categories according to the noise level, remove the suspicious components in the category with a large mean noise level, and implement denoising.

[0076] It should be noted that experimental equipment sometimes has abnormal situations, among which the probability of abnormal situations is relatively small, and the intensity of some abnormal situations is relatively small. Through the above regular change analysis, it is also easy to filter out the information in abnormal situations. In order to prevent denoising from removing abnormal information, it is necessary to further analyze the abnormal information and noise information.

[0077] S30: Calculate the noise level of the suspicious component.

[0078] It should be noted that when an abnormality occurs in the experimental equipment, it is reflected in the data of multiple dimensions, so the correlation of abnormal data should be relatively large. For example, when the power of the experimental equipment is abnormal, the temperature will also be abnormal, so the correlation of abnormal data of different dimensions of the experimental equipment is relatively small. Noise generally only affects data of a single dimension. For example, when the sensor is interfered by thermal noise or electromagnetic noise when measuring current, noise exists in the measured current data. This noise is only caused by sensor factors, so the correlation between noises of different dimensions is small.

[0079] It needs to be further explained that the regularity of noise is generally smaller, while the regularity of anomalies is generally larger. For example, when the temperature of experimental equipment exceeds the temperature limit, the operation of the equipment will become abnormal. Therefore, this abnormality has a certain regularity, which is higher than the regularity of noise. Therefore, noise and anomalies can be distinguished by analyzing the regularity.

[0080] Preferably, as an example, calculating the noise level of the suspicious component includes:

[0081]

[0082] Among them, X represents the mean correlation between the suspicious component and the experimental equipment time series of all other dimensions, B represents the distribution interval difference between the newly added peaks obtained after superimposing the new suspicious component, T represents the morphological difference between the newly added peaks obtained after superimposing the new suspicious component, and Z represents the noise level of the suspicious component.

[0083] It can be understood that the greater the correlation between the suspicious component and the experimental equipment time series of all other dimensions, the less likely the suspicious component is noise. Among them, X reflects the correlation between the suspicious component and the experimental equipment time series of all other dimensions. The larger the value, the less likely the suspicious component is noise. After the suspicious component is added, the regularity of the changed part of the signal is greater, indicating that the suspicious component is more likely to be abnormal. Among them, B reflects the distribution regularity of the changed part of the signal after the suspicious component is added. The larger the value, the weaker the distribution regularity of the changed part of the signal after the suspicious component is added, and the greater the possibility that the suspicious component is noise. T reflects the change regularity of the changed part of the signal after the suspicious component is added. The larger the value, the weaker the change regularity of the changed part of the signal after the suspicious component is added, and the greater the possibility that the suspicious component is noise.

[0084] The above embodiments involve distribution interval differences and morphological differences. The following describes methods for obtaining distribution interval differences and morphological differences.

[0085] First, the method of obtaining the distribution interval difference is introduced.

[0086] Preferably, as an example, the method for obtaining the distribution interval difference includes:

[0087] The horizontal distance between the extreme point of the newly added peak / valley and the extreme point of the previous newly added peak / valley is obtained as the distribution interval of the newly added peak / valley, and the variance of the distribution intervals of all the newly added peaks / valleys is taken as the distribution interval difference.

[0088] It should be noted that, if the above processing is for a newly added peak, the horizontal distance between the newly added peak and the previous newly added peak is used as the distribution interval of the newly added peak; if the above processing is for a newly added valley, the horizontal distance between the newly added valley and the previous newly added valley is used as the distribution interval of the new valley.

[0089] It can be understood that the distribution interval of the newly added peaks / valleys reflects the temporal distribution law of the signal change portion caused by the addition of the suspicious component. The larger the value, the weaker the temporal distribution law of the signal change portion caused by the addition of the suspicious component.

[0090] Then the method of obtaining morphological differences is introduced.

[0091] Preferably, as an example, the method for obtaining the morphological difference includes:

[0092] The cosine similarity between the newly added peak / valley and the previous newly added peak / valley is obtained as the adjacent similarity of the newly added peak / valley, and the average of the adjacent similarities of all the newly added peaks / valleys is taken as the morphological difference.

[0093] It should be noted that, if the above processing is for a newly added peak, the cosine similarity between the newly added peak and the previous newly added peak is used as the adjacent similarity of the newly added peak; if the above processing is for a newly added valley, the cosine similarity between the newly added valley and the previous newly added valley is used as the adjacent similarity of the newly added valley.

[0094] It can be understood that the distribution interval of the newly added peaks / valleys reflects the variation pattern of the signal change portion caused by the addition of suspicious components. The larger the value, the weaker the variation pattern of the signal change portion caused by the addition of suspicious components.

[0095] S31: clustering suspicious components into two categories according to the noise level, removing suspicious components in the category with a large mean noise level, and implementing denoising.

[0096] Preferably, as an example, suspicious components are clustered into two categories according to the noise level, and suspicious components in the category with a large mean noise level are removed to implement denoising, including:

[0097] According to the noise level, the K-means algorithm is used to cluster the suspicious components into two categories. The mean noise level of all suspicious components in each category is calculated. The suspicious components in the category with the mean noise level are removed, and all the remaining frequency components are superimposed together to obtain the denoised experimental equipment data time series.

[0098] At this point, this embodiment is completed.

[0099] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. An AI-based experimental equipment data processing method, characterized in that: include: Obtain the experimental equipment data time series sequence of each dimension, use sensors to collect the data of each dimension of the experimental equipment at each moment, and arrange the data of all moments of each dimension in time series to obtain the experimental equipment data time series sequence of each dimension; the experimental equipment is a temperature controller, and the dimension types are current, voltage, and temperature; Obtain the spectrum of the experimental equipment data time series, arrange all frequency components in the spectrum in ascending order according to the frequency size, superimpose the frequency components in sequence, and calculate the regularity change amount after each superposition of a new frequency component , obtain the newly added peaks / valleys after superimposing the new frequency components and the monotonic segment where each newly added peak / valley is located, including: The sequence after the new frequency component is superimposed is recorded as the post-superimposition sequence, and the sequence before the new frequency component is superimposed is recorded as the pre-superimposition sequence; Get all extreme values ​​in the sequence before superposition, and get all extreme values ​​in the sequence after superposition; Compared with the sequence before superposition, the extreme value that only exists in the sequence after superposition is taken as the target extreme value; Get the extreme values ​​on both sides of the target extreme value, and use the area between the extreme values ​​on both sides of the superimposed sequence as the newly added peak / valley; Use extreme points to divide the sequence before superposition into several monotonic segments; obtain the time period corresponding to the monotonic segment, obtain the time period corresponding to the newly added peak / valley, and use the monotonic segment that intersects with the time period of the newly added peak / valley as the monotonic segment where the newly added peak / valley is located; Indicates the first The slope at the data point, Indicates the number of data in the experimental equipment data time series sequence, , Respectively represent the maximum fluctuation and span of the jth newly added peak / valley, represents the length of the monotonic segment where the jth newly added peak / valley is located, M represents the number of newly added peaks / valleys, G represents the regular change amount, norm() represents linear normalization processing, and a number of suspicious components are screened out by comparing the regular change amount with the preset change amount threshold; The method for obtaining the maximum fluctuation amount includes: Obtain two boundary points of the newly added peak / valley, and take the mean of the extreme value in the newly added peak / valley and the longitudinal distance between the two boundary points as the maximum fluctuation of the newly added peak / valley; Calculate the noise level of the suspected component , X represents the mean correlation between the suspicious component and the experimental equipment time series of all other dimensions, They respectively represent the distribution interval difference and morphological difference between the newly added peaks / valleys obtained after superimposing new suspicious components. The suspicious components are clustered into two categories according to the noise level, and the suspicious components in the category with a large mean noise level are removed to achieve denoising.

2. According to the AI-based experimental equipment data processing method of claim 1, it is characterized in that: The sequentially superimposing frequency components includes: First, the first frequency component is taken as the reference sequence, and the sequence obtained by superimposing the second frequency component and the reference sequence is recorded as a new reference sequence; then the sequence obtained by superimposing the third frequency component and the new reference sequence is recorded as a new reference sequence, and so on, until all frequency components are superimposed.

3. The AI-based experimental equipment data processing method according to claim 1, characterized in that: The method for obtaining the span includes: The two boundary points of the newly added peak / valley are obtained, and the horizontal distance between the two boundary points of the newly added peak / valley is used as the span of the newly added peak / valley.

4. The AI-based experimental equipment data processing method according to claim 1, characterized in that: The method for obtaining the distribution interval difference includes: The horizontal distance between the extreme point of the newly added peak / valley and the extreme point of the previous newly added peak / valley is obtained as the distribution interval of the newly added peak / valley, and the variance of the distribution intervals of all the newly added peaks / valleys is taken as the distribution interval difference.

5. The AI-based experimental equipment data processing method according to claim 1, characterized in that: The method for obtaining the morphological difference comprises: The cosine similarity between the newly added peak / valley and the previous newly added peak / valley is obtained as the adjacent similarity of the newly added peak / valley, and the average of the adjacent similarities of all the newly added peaks / valleys is taken as the morphological difference.

6. The AI-based experimental equipment data processing method according to claim 1, characterized in that: The method of removing suspicious components in the category with a large noise level mean value to implement denoising includes: The mean of the noise levels of all suspicious components in each category is calculated, the suspicious components in the category with the mean noise level are removed, and all the remaining frequency components are superimposed together to obtain the denoised experimental equipment data time series.

Images